Craniofacial Fractures Studies On Association Of Midface And Lower Face With Frontal Bone Injuries Using Integration Of Multilayer Perceptron (Mlp) And Logit Model Approach

Main Article Content

Wan Muhamad Amir W Ahmad, Tang Liszen, Nor Azlida Aleng, Noraini Mohamad, Mohd Faizal Abdullah, Nur Mohamad Mohd Makhatar, Mohamad Shafiq Mohd Ibrahim, Nor Farid Mohd Noor, Farah Muna Mohamad Ghazali, Mohamad Nasarudin Adnan

Abstract

Introduction: The number of patients who present with facial injuries every year is on the rise. Most admission requires combined intervention by neurosurgery and maxillofacial team due to frontal bone fractures associated with various types of brain injury. The most common form of skull bone fracture is a frontal bone fracture. A high-impact head injury can fracture the frontal bone and other nearby bones. Objectives and Method: There is a retrospective study of patients with maxillofacial trauma at Hospital Universiti Sains Malaysia (USM) over five years (1 January 2012 to 31 December 2016). The hospital records of patients who sustained these fractures were analyzed using the newly developed R syntax. This study aims to determine which facial bone fractures are associated with a frontal bone fracture in maxillofacial trauma that occurs at the same time. Therefore, this study proposes an application of Artificial Neural Networks (ANNs) through a feed-forward network toward clinical study data on craniofacial fractures. The most associated bones related to the frontal bone fracture will be determined and will be the input for the multiple logistic regression (MLR). The analysis will be conducted entirely using developed R syntax. The generated syntax is divided into three major sections: Bootstrap (B), Multilayer Perceptron (MLP), and Multiple Logistic Regression. Results: This type of fracture occurred in 218 patients, with 80.7% male and 19.3% female. There is four variable which was Gender ( ), Le Fort III fracture ( ), mandibular symphysis fracture ( ), and mandibular condylar fracture ( ). The above MLP gave the lowest mean absolute deviance (0.0007179404). The accuracy obtained is about 99.928%. Conclusions: A Multilayer Feed-Forward Neural Network (MLFF) with multiple logistics regression for the modeling and prediction purpose of collected data is a good approach. The result obtained is being tested and checked from an important clinical point of view. This approachable technique was discovered to have superiority in the variable selection for multiple logistic regression modeling. In real life, many of the relationships between inputs and outputs are non-linear as well as complex relationships.  As a result, using MLFF for variable selection, especially for modeling purposes, is a very good strategy and was discovered to have superiority of the variable selection for multiple logistic regression modeling.

Article Details

Section
Articles